Platform Traffic Costing More? It Isn't Just Higher Ad Bids — Buyers Are Now Vetting Suppliers with AI

B2B platform traffic keeps getting more expensive, and most people blame rising bids and more competitors. A bigger change goes unnoticed: where buyers begin sourcing is itself shifting. Alibaba.com's own disclosed figures confirm it.

Let's be precise. Traffic is genuinely more expensive, but there are at least two reasons. Everyone knows the first — more people competing for the same slot. The second is the layer AI search optimization deals with, and most people haven't caught on yet.

Platforms cost more — but not for the same reason

Almost everyone in export trade feels the same thing: the platform costs more every year. That is also why AI search optimization keeps coming up. Membership fees are rising, paid-ad click costs are rising, and inquiry quality is falling.

Some of the numbers here are hard numbers — from a listed company's annual report, not an industry estimate.

Alibaba Group's fiscal year 2026 annual report (year ended 31 March 2026) contains two figures: Alibaba.com has more than 240,000 paid supplier members in China and worldwide, and over the same period more than 47 million buyers from more than 190 countries came to it to find opportunities or complete transactions.

240,000 suppliers share one buyer pool. The slots where buyers can actually see you are limited. That is the first reason traffic costs more — more people competing for the same slots, while the number of slots has not grown.

The same report carries a line worth pausing on: in fiscal 2026, value-added services and customer management services contributed the majority of Alibaba.com's total revenue.

In other words, the platform's largest revenue line is no longer membership fees but value-added and customer management services, including P4P advertising. Put plainly, the platform's main income is what merchants pay for visibility. That sentence says a great deal about where the bargaining power sits between buyer and seller.

But all of this is the old problem. GEO is about the second one.

Buyers have stopped typing keywords — and the platform admits it

On 9 September 2026, Alibaba.com's annual overseas buyer event, CoCreate, opened in Los Angeles. More than 15,000 US small and medium businesses attended — the year before, that number was 3,000.

On the main stage, Zhang Kuo, President of Alibaba.com, said: "The future of B2B is A2A — agent to agent."

This is not a forecast; it is already happening. According to Economic Daily (Jingji Ribao), the behaviour of overseas buyers on Alibaba.com has shifted structurally: they no longer type keywords into a search box. They describe what they need in plain language, and AI then handles spotting trends, refining products, screening suppliers, negotiating terms and managing order workflows.

The accompanying figures are blunt:

• Overseas buyers' use of AI multimodal search has driven traffic growth of more than 132%
• Global users of Accio, Alibaba.com's AI sourcing workspace, grew more than 30-fold in the past year
• 10 million SMEs now do business through Accio
• AI token consumption on Accio grew 6-fold in four months

One distinction matters here: these are platform-side usage figures. They do not mean your inquiries will grow. The 132% is traffic growth — not inquiry growth, and certainly not order growth. Between buyers using AI to source more often and your orders going up sits one more link: whether AI can read you.

But directionally it shows one thing: "let AI do a first round of screening, then decide who to contact" is already part of how overseas buyers source, every day.

For anyone doing AI search optimization, the meaning is direct: the sourcing entry point has moved from the search box to the chat box.

The rules changed: from "which page do I rank on" to "am I on the shortlist"

Two actions, side by side, explain it completely.

How buyers used to find suppliers: open the search box, type keywords, scroll the list, compare prices one by one, send an inquiry. In that flow, what you optimised was rank — the higher you sat, the more likely you were to be clicked.

What more and more buyers do now: describe what they need, AI interprets it, AI returns a shortlist of suppliers, the buyer verifies them, then makes contact. In that flow, the buyer may never open your store page at all.

This is why three things happen at once: you spend more, impressions rise, and inquiry quality falls. You are raising your bets under the old rules while buyers are already picking people under the new ones. This is also where GEO divides from traditional SEO.

This is not unique to Alibaba.com

Someone will say: that is the platform pushing its own AI; elsewhere it may not apply. But the logic is not unique to Alibaba.com — it is the same reason GEO exists at all.

Independent research points the same way. Pew Research Center published a behavioural study in July 2025 tracking 900 US adults across 68,879 Google searches (March 2025):

• When an AI summary appeared on the page, 8% of users clicked a traditional search result; without a summary, 15% did
• Only 1% clicked through to a source link cited inside the summary
• When a summary appeared, 26% of visits ended on the results page itself (16% without one)

(This is an observational study, not a randomised experiment, so it cannot be read as direct causation — but the direction is clear enough.)

Two more pieces of evidence point the same way: an August 2026 preprint reporting a randomised experiment found that being assigned to an AI search mode cut outbound click-through by 18.8 percentage points; and a study of Wikipedia estimated that AI summaries reduced search traffic to English entries by roughly 15%.

Put those together with the platform figures and the conclusion is this: "read the AI answer first, then decide whether to click into any website" is becoming a general way of getting information. Sourcing is simply how it shows up in a B2B setting.

This is exactly the premise behind why Generative Engine Optimization keeps coming up: the entry point changed, and so did the basis for being chosen.

240,000 suppliers competing for one AI shortlist — what gets you picked

Back to that number: 240,000 paid suppliers.

Buyers no longer page through listings. They get a shortlist from AI. So what does AI use to put you on it?

Not storefront design, not ad budget — but facts it can read and verify: minimum order quantity, sample lead time, monthly capacity, which materials you work in, your certification numbers, what cases you have delivered.

Now look at your own product pages and store details. Do they say "supports all kinds of customisation", "reliable quality", "on-time delivery", "strong capabilities" — or do they carry the numbers that can actually be checked?

When AI cannot find numbers on your page, it names the competitors who wrote theirs down. And the shortlist has a fixed number of places — if you are not on it, the buyer never knows you existed. That is the core problem GEO addresses.

What to do on Alibaba.com, and what to do off it

One thing needs to be stated clearly, because plenty of owners understand it backwards.

GEO — Generative Engine Optimization, what people usually call AI search optimization — is not about leaving the platform. It is about making AI able to read you. And it has two ends.

The first end is on-platform. Alibaba.com is already AI-ifying itself, and that half is precisely where AI search optimization follows on the platform side. It means whether your store data and product information are structured and complete directly determines whether AI can read you in. Fill in attributes, specifications, certifications, MOQ and lead time completely and accurately — that step can be done right in the Alibaba.com back office.

The second end is off-platform. Buyers do not only ask AI inside the platform. They ask English questions on ChatGPT and Perplexity, and Chinese questions on Doubao, DeepSeek, ERNIE Bot, Kimi, Tencent Yuanbao and Qwen. Those engines read public web pages and content platforms — your own website, media coverage, discussions in industry communities. The platform cannot manage this end for you; you have to build it.

The two ends are not a choice; they are two positions of the same job. Do only the on-platform half and you may already be off the shortlist when a buyer asks AI off-platform. Do only the off-platform half and your on-platform information is too thin for AI to conclude "this company can do it".

There is one more easily missed detail, and it is the single most important link in generative engine optimization: AI cross-checks across sources. Your website states one founding year, your platform profile states another, an article states a third — when they do not line up, AI would rather not use you. But this does not mean pasting the same paragraph everywhere. What is required is consistency of facts, not identical wording. Reposting text verbatim can be flagged as duplicate content and actually lower your citation rating.

Shenzhen Yihang Network Technology Co., Ltd. expanded into GEO in 2024, after years of working on Alibaba.com-related services — which is why these two ends connect in our practice: one is how to write the information in your store, the other is how to build public sources off the platform. We have served a good number of export businesses in the Pearl River Delta and the Yangtze River Delta. Our method is published at www.chyihang.com/en/geo/ — go and have a look.

On "customer acquisition costs rose by X percent" — why I won't give you a number

Before writing this, I went through the literature. I had intended to give you a "customer acquisition costs for export businesses rose XX% year on year".

After checking, I decided not to write it.

For the same year, different sources give 35%, 37%, 42%, 43%, and 67%. Some say B2B platform acquisition costs doubled in three years; others say the cost per customer at trade shows rose 60% to 80%. Trace them back and most turn out to be vendors' own promotional copy, or "a recap after talking to clients" — not statistics anyone can verify.

A set of contradictory numbers does not help you decide anything. It only helps someone sell something.

What can be confirmed is structure, not percentage growth: paid suppliers compete in one pool (240,000, from the annual report), and buyers' sourcing entry point is migrating towards AI (132%, 30-fold, 10 million, from platform disclosures and state-media reporting). The first says competition has not become easier. The second says the way you compete has changed.

As for how much acquisition costs have actually risen in your own category — only your own back office can tell you that. Anyone else's number is worth nothing.

Three questions about platforms and AI that need straight answers

The platform is pushing its own AI tools. Is filling in my store information completely enough?

Not enough — but it is a necessary first step. On-platform structuring only solves "can AI read me inside the platform". When a buyer asks AI off-platform, it reads public web pages. You need both ends for a stable result.

And there is a practical difference: platform rules are set by the platform. Off-platform public sources are your own asset.

How long before I see a change?

This is where AI search optimization is most often misunderstood: it solves "being read", not "ranking first".

No promises on timeline, no promises on outcome. Any claim that you will "be in AI recommendations within 90 days" lacks verifiable support. The openly observed pattern: on time-sensitive platforms, content weight concentrates in the first week or two; on some platforms it takes about three months; evergreen content keeps getting cited. This is an ongoing activity, not one batch of articles and done.

Is it too late to start?

The data says AI sourcing is still in a rapid-penetration phase — platform-side AI search traffic grew 132% in a year and Accio users grew 30-fold. That means "how you get seen" has not settled yet.

Where nothing has settled, there is room to take a position. The GEO window is essentially this unsettled phase. Wait until the rules are fully locked and the barrier to entry only gets higher.

A closing note

Expensive platform traffic is two reasons stacked: more people competing for the same slot, and — the rules that decide whether you are seen have been replaced.

You cannot change the first. In the second, half the work is yours to do: write down the facts that are already true about you in a form AI can read, verify and cite.

Generative Engine Optimization, at the end of it, does one thing: it puts facts in the right place. That is not the same as switching channels.

When did you last search for your own product in AI? Who came up? Tell us in the comments.

References

1. Economic Daily (Jingji Ribao), "Alibaba.com CoCreate Summit Held in the United States," 10 September 2026 (contains remarks by Zhang Kuo, President of Alibaba.com, and data disclosed by the platform). jingjiribao.cn
2. Alibaba Group, Fiscal Year 2026 Annual Report, 18 June 2026 (disclosures on Alibaba.com paid supplier members, buyer numbers, revenue mix and Accio). hkexnews.hk
3. Pew Research Center, "Google users are less likely to click on links when an AI summary appears in the results," 22 July 2025 (sample: 900 US adults, 68,879 Google searches). pewresearch.org

Data and scope note

The platform figures cited here come from Alibaba.com's public disclosures and media reporting; the annual report figures come from a listed company's public disclosures. They are company- or platform-level statistics, used to illustrate structural change in sourcing behaviour and information-seeking. They do not mean that any specific industry or individual company will necessarily receive inquiries or orders. Descriptions of how each AI platform sources and recommends content come from public material and industry observation; no platform publishes its recommendation mechanism officially, and behaviour may change as platforms iterate. Before acting on this, a company should still assess its own products, target markets and content base independently.

#Alibaba.com #Export #AISearch #GEO